I am an Assistant Research Scientist at University of Michigan Transportation Research Institute (UMTRI). Before joining UMTRI, I was a post-doctoral researcher at the Michigan Traffic Lab of the University of Michigan, Ann Arbor, working with Prof. Henry X. Liu. I received the bachelor’s and Ph.D. degrees in the Department of Automation at Tsinghua University, China, in 2014 and 2019, respectively, where I was advised by Prof. Yi Zhang. I was also a visiting Ph.D. student in Civil and Environmental Engineering with the University of Michigan from 2017 to 2019, where I was advised by Prof. Henry X. Liu.
I received the Best Ph.D. Dissertation Award from the IEEE Intelligent Transportation Systems Society (ITSS) in 2020. I am serving various roles in multiple academic organizations, including Member in the SAE On-Road Automated Driving (ORAD) Verification and Validation Task Force, Organizer of the IEEE 2021 Intelligent Vehicles Symposium “Cooperative Driving in Mixed Traffic” Workshop, and Organizer of the 38th International Conference on Machine Learning “A Blessing in Disguise: The Prospects and Perils of Adversarial Machine Learning” Workshop.
My current research interests lie in developing and validating safety-critical machine learning, particularly for connected and automated vehicle testing and evaluation. On these topics, I have published around 20 articles on refereed journals, including Nature Communications, IEEE Trans. Intell. Transp. Syst., IEEE Trans. Control Syst. Technol., and Transp. Res. Part C Emerg. Technol..
Ph.D. in Control Science and Engineering, 2019
Visiting Scholar in Civil and Environmental Engineering, 2017-2019
University of Michigan, Ann Arbor
B.S. in Automation, 2014
For the complete list of my publication, please see my Google Scholar page.
This paper proposes an optimal and computationally efficient cooperative driving strategy with the polynomial-time complexity. By modeling the conflict relations among the vehicles, the solution space of the cooperative driving problem is completely represented by a newly designed small-size state space. Then, based on dynamic programming, the globally optimal solution can be searched inside the state space efficiently. It is proved that the proposed strategy can reduce the time complexity of computation from exponential to a small-degree polynomial.
This paper presents a learning-based stochastic driving model that meets the unique needs of AV testing (i.e., interactive and human-like stochasticity). The model is built based on the long short-term memory architecture. By incorporating the concept of quantile regression into the loss function of the model, the stochastic behaviors are reproduced without prior assumption of human drivers.
In this paper, a unified framework is proposed to generate corner cases for decision-making systems. To address the challenge brought by high dimensionality, the driving environment is formulated based on the Markov decision process, and the deep reinforcement learning techniques are applied to learn the behavior policy of BVs. With the learned policy, BVs behave and interact with the CAVs more aggressively, resulting in more corner cases.
A robust platoon control framework is proposed for mixed traffic flow where connected and automated vehicles (CAVs) and human-driven vehicles (HDVs) coexist. The prediction uncertainty is dynamically mitigated by the feedback control and restricted inside a set with a high probability. When the uncertainty exceeds the set or additional external disturbance emerges, the feedforward control is triggered to plan a “tube” (a sequence of sets), which can bound CAVs' actual trajectories. As the replanning process is usually not required, the proposed method is much more efficient regarding computation and communication, compared with the MPC method.
Tests for autonomous vehicles are usually made in the naturalistic driving environment where safety-critical scenarios are rare. Feng et al. propose a testing approach combining naturalistic and adversarial environment which allows to accelerate testing process and detect dangerous driving events.
How to generate testing scenario libraries for connected and automated vehicles (CAVs) is a major challenge faced by the industry. In previous studies, to evaluate maneuver challenge of a scenario, surrogate models (SMs) are often used without explicit knowledge of the CAV under test. However, performance dissimilarities between the SM and the CAV under test usually exist, and it can lead to the generation of suboptimal scenario libraries. In this article, an adaptive testing scenario library generation (ATSLG) method is proposed to solve this problem.
This paper presents a new safety assessment framework for highly automated driving systems in test tracks. The framework integrates an augmented reality testing platform and a testing scenario library generation method together. The framework has been implemented in Mcity test facility with a SAE Level-4 ADS vehicle. The framework can accelerate the assessment process by multiple orders of magnitude comparing to the on-road test approach.
This paper aims to provide implementation examples and guidelines, and to enhance the proposed methodology under high-dimensional scenarios. Three typical cases, including cut-in, highway-exit, and car-following, are designed and studied in this paper.
Testing and evaluation is a critical step in the development and deployment of connected and automated vehicles (CAVs), and yet there is no systematic framework to generate testing scenario library. This study aims to provide a general framework for the testing scenario library generation (TSLG) problem with different operational design domains (ODDs), CAV models, and performance metrics.
A computationally efficient strategy is proposed to obtain the globally optimal passing order based on dynamic programming (DP). Specifically, the problem of merging at on-ramps is resolved by a DP method, which uses the domain knowledge to reduce the complexity by well defining the state space, state transition, and criterion function. With the DP method, it is proved that the globally optimal passing order can be obtained with the quadratic polynomial computational complexity of O(N^2), where N denotes the number of vehicles.
This paper aims to clarify the relationship of ambiguous definitions and various analysis methods of string stability for vehicular platoon control, providing a rigorous foundation for future studies. A series of equivalences are summarized and discussed. The pros and cons of different analysis methods and definitions are discussed, too.
Most existing methods for string stability are laborious for implementation without considering either heterogeneous disturbances (e.g., tracking errors and unmodeled dynamics) or saturation constraints (e.g., input saturation). We use feedforward control for large yet infrequent disturbances and feedback control for small yet frequent disturbances. Different from MPC, our feedforward control is event-triggered so that the intervehicle communication and planning costs can be significantly reduced. Different from pure robust feedback control, our combination of feedback and feedforward control could reduce the conservation of the controller.
Both the daily and long-range temporal dependence exert considerable influence on the traffic flow series. The daily temporal dependence creates crossover phenomenon when estimating the Hurst. PCA-based method turns out to be a better method to extract the daily temporal dependence.
In this study, the authors compare the influence of different data reformatting, archive the bursts defined before in descending order with respect to the absolute value of the burst points and propose a flexible compression framework to balance between burst components and low-dimensional intra-day trend.